Skip to content
Sign in Create account

Reading text in images

Optical character recognition (OCR) reads text in screenshots, photos and other supported images on your device. Review the extracted text before sending. OCR can miss text, including names, accents and whole lines.

Language selection

The browser extension, browser portal, CloakScan, and Cloakup select English plus supported languages from your browser preferences. Desktop chat uses the languages its webview reports from your computer. Norwegian Bokmål, Nynorsk and Norwegian all select the Norwegian pack.

English is available by default. Other packs download in the background from CloakAPI. If a pack is still downloading or unavailable, the app reads with the packs it has and shows which it used. Use the reading-language settings to choose specific packs or return to automatic selection. Where the review offers another language, you can request another read with that pack.

Where image description is available, it uses a separate local model.

Local integrations

  • Native SDKs. The native engine uses Tesseract with English by default. To use other languages, install their Tesseract packs and set CLOAK_OCR_LANGS (for example, eng+nor). C ABI callers can pass ocr_langs to select installed packs for a file.
  • TypeScript browser SDK. The browser module accepts extractedText from your own on-device image reader. Its distribution does not include a Tesseract worker, language packs, or a trusted asset host. Your app selects and prepares its local reader before calling the SDK.
  • MCP server. Starting in 0.6.2, image OCR requests English and supported process-locale languages. Install the Tesseract packs you need locally. Set ocr_langs in the tool call or CLOAK_OCR_LANGS for the process to choose specific packs. When the tool reads the image, it reports the used and unavailable packs. The MCP host implements review and any later sending.
  • Local proxy. Starting in 0.4.1, the proxy requests English and supported process-locale languages. It prepares missing packs in the background and uses the available requested packs for each image, with English as a fallback. The local log records which languages were selected. Set CLOAK_OCR_LANGS=eng+nor to override the default. Use an application that can present and approve the exact prepared content before sending.

CloakConvert, CloakCV, CloakDocs, CloakMeet, CloakPost, CloakRedact, CloakSheets, CloakTranslate, and CloakWrite have no image-to-text OCR entry point. Open the image in an app that can read it locally.

See File attachments for supported file paths and review requirements.